mxochicale / Yolov5-inference-on-ONNXRuntime-and-opencv-DNN

Yolov5-inference-on-ONNXRuntime-and-Opencv-DNN

• To convert yolov5 weights to onnx format follow this tutorial link: ultralytics/yolov5#251.

• yolov5.onnx model in Netron

ONNXRuntime

• For GPU system install ONNXRuntime-GPU library and ONNXRuntime for CPU system.

!pip install -r requirements.txt

alternatively, using mamba

mamba update -n base mamba
mamba create -n yolocvVE pip -c conda-forge
mamba activate yolocvVE
pip install -r requirements.txt

• Command to run the code:

python yolov5_onnxinfer.py  --image ./bus.jpg  --weights ./yolov5s.onnx --conf_thres 0.7 --iou_thres 0.5 --imgs 640 --classes ./classes.txt

• Arguments Details:

1. Input image
2. Weight file
3. Confidence threshold value
4. IOU threshold value
5. Image size
6. Classes.txt file

Opencv DNN

• Command to run code:

python Yolov5_infer_opencv.py --image ./bus.jpg --weights ./yolov5s.onnx --classes ./classes.txt --imgs_w 640 --imgs_h 640 --conf_thres 0.7 --score_thres 0.5 --nms_thres 0.5

• Arguments Details:

1. Input image
2. Weight file
3. Image width
4. Image Height
5. Confidence threshold value
6. score threshold value
7. nms threshold value
8. Classes.txt file

• Comparison of inference time: For image 'bus.jpg', inference time of ONNXRuntime and opencv DNN module are:

  1. ONNXRuntime: 0.16300702095031738
  2. opencv DNN: 0.4327085018157959

Streamlit yolov5 Inference App

• Run stramlit code.

streamlit run Streamlit_yolov5_infer.py

• References

1. https://github.com/ultralytics/yolov5
2. https://github.com/WongKinYiu/yolov7
3. https://learnopencv.com/object-detection-using-yolov5-and-opencv-dnn-in-c-and-python/